Speech enhancement method and system

Abstract

The invention discloses a speech enhancement method and system. The method comprises the following steps: acquiring a noisy speech signal; performing wavelet decomposition on the noisy voice signal to obtain a plurality of noisy sub-bands; inputting each sub-band with noise into a speech enhancement model to obtain an enhanced sub-band corresponding to each sub-band with noise; and performing wavelet synthesis on the plurality of enhancement sub-bands to obtain an enhanced voice signal. According to the invention, the length of the signal can be reduced layer by layer through discrete wavelet change, the number of sampling points is reduced, the method is more suitable for non-stationary signals such as voice, and the voice signal enhancement effect is improved.

Description

technical field [0001] The invention relates to the technical field of audio processing, in particular to a voice enhancement method and system. Background technique [0002] In practical applications, speech signals are easily disturbed by noise, and it is necessary to suppress noise interference through speech enhancement technology, reduce the impact of noise on speech, and extract useful speech signals from noisy speech. The current speech enhancement technology is mainly a speech enhancement method based on deep learning, that is, two kinds of audio features are used as the input of the network, one of which is based on the audio time-domain waveform for speech enhancement, and the other method is to perform speech enhancement first. Some signal preprocessing means such as short-time Fourier transform, and then do noise reduction processing. [0003] However, in the way of speech enhancement based on the audio time-domain waveform, due to the dense sampling points of t...

AI Technical Summary

Problems solved by technology

[0003] However, in the way of speech enhancement based on the audio time-domain waveform, due to the dense sampling points of the time-domain signal, for long audio, it is difficult for the network to learn all the information of the entire audio, so it is necessary to divide the signal into frames, so that the network can Frame learning, and finally splicing each frame together, which will cause serious distortion at the audio splicing and poor enhancement effect

However, based on short-time Fourier transform processing, the signal needs to be a stationary signal. For non-stationary signals, the frequency components of the signal are different at different times, which cannot be distinguished. Therefore, the scope of application of this method is limited. The enhancement effect of stationary signal is poor

Images